Skip to documentation
On this page

Use with AI

Bring an idea from your favorite agent into an editable design.

Start with a prompt

Describe your idea to your favorite agent, then keep shaping the result in GradientArt. This works with web agents such as ChatGPT and local agents such as Copilot. No plugin, repository checkout, or GradientArt account is needed to open or import a design.

Paste into your agent, then replace the description in brackets with your idea.

Your agent creates the artwork; GradientArt validates it and opens it as editable layers. A documentation link does not grant an agent API access. What happens next depends on the tools your agent has.

Web agents, including ChatGPT

Paste the prompt into your chat. If the agent can call HTTP APIs, it can return an Open in GradientArt link. Browsing a website alone is not enough to make the required POST request.

If it cannot call the API, ask for a .gradient-art.json file instead. Open the editor, choose Project actions > Import JSON, and select the file. If the agent gives you a code block, save only the JSON as design.gradient-art.json first. See Import an editable JSON file.

If your agent cannot browse the guide, download and attach the authoring guide, self-contained schema, and minimal example to the conversation. Do not ask it to guess the file format.

A dedicated ChatGPT GPT

A configured GPT can call the preparation API through an Action and return the editor link directly. There is no published GradientArt GPT linked here yet. The GPT setup kit includes ready-to-paste instructions, knowledge files, and a self-contained Action definition.

Creating and publishing a GPT depends on your ChatGPT account and workspace. OpenAI currently restricts new GPT creation on personal accounts and has announced a move toward plugins. Check OpenAI's current availability guidance before setting one up. The universal prompt and file-import route do not require a custom GPT.

Connect a ChatGPT plugin with MCP

If your plugin can generate designs but cannot make the POST request, connect the hosted MCP server at https://gra.dient.art/api/mcp using Streamable HTTP with no authentication. It exposes one tool, prepare_design, which accepts the complete versioned file in its document argument.

The MCP connection guide includes a prompt for updating your existing plugin without changing its identity or private visibility. The endpoint is publicly callable even if your plugin stays private.

The tool returns the same one-hour editor links and uses the same validation and quotas as the HTTP API. It does not save projects. Plugin availability and connection permissions depend on your account.

Copilot and local agents

Paste the same prompt into Copilot or another agent with terminal access. Allow it to read the public guide and, when asked, make the HTTP request. No MCP server, API key, or local GradientArt installation is required.

After writing design.gradient-art.json, the agent can submit it with:

curl --fail-with-body \
  -H 'Content-Type: application/json' \
  --data-binary @design.gradient-art.json \
  https://gra.dient.art/api/v1/designs/prepare

The response contains editorUrl, expiresAt, and saved: false. Ask the agent to return that exact URL, not a URL containing the design JSON. Small and large designs use the same temporary-link API.

Open, edit, and keep your design

Links expire one hour after creation. Anyone with the link can access the original generated design until expiry, so keep it private and leave sensitive information out of your design. The API temporarily stores the design; it does not create a saved project.

Opening a design does not save it. Choose Save to create an online project (sign-in required), or Export > JSON project to keep a local editable file. Save or export before reloading or closing the editor.

If you already have work in the editor, you must confirm replacement. Cancel keeps that work intact. Reopening the original link restores the agent's design, not edits you made afterward.

When something goes wrong

  • The link expired: Ask the agent to prepare a new link from its original JSON.
  • The API rejected the design: Give the structured error back to the agent so it can repair the indicated fields. Schema validation alone does not check all clone relationships, formulas, and design limits.
  • The API is rate-limited or unavailable: Respect Retry-After and retry with backoff, or ask for a JSON file to import. Never treat an error as a prepared link.
  • A ChatGPT Action cannot send a large design: Actions have a smaller payload limit than this API. Use the JSON-file route or a terminal-capable agent.

Integration reference

Use the agent guide for generation rules, quotas, and retention details. The modular schema lets agents read only the layer types they need; the OpenAPI definition describes the HTTP contract. For MCP-compatible clients, use the remote MCP connection guide.